Author ORCID Identifier

https://orcid.org/0000-0002-1606-4856

Date of Award

Summer 8-9-2022

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Juan M. Banda

Abstract

Over the last decade, advances in machine learning have led to an exponential growth in artificial intelligence i.e., machine learning models capable of learning from vast amounts of data to perform several tasks such as text classification, regression, machine translation, speech recognition, and many others. While massive volumes of data are available, due to the manual curation process involved in the generation of training datasets, only a percentage of the data is used to train machine learning models. The process of labeling data with a ground-truth value is extremely tedious, expensive, and is the major bottleneck of supervised learning. To curtail this, the theory of noisy learning can be employed where data labeled through heuristics, knowledge bases and weak classifiers can be utilized for training, instead of data obtained through manual annotation. The assumption here is that a large volume of training data, which contains noise and acquired through an automated process, can compensate for the lack of manual labels. In this study, we utilize heuristic based approaches to create noisy silver standard datasets. We extensively tested the theory of noisy learning on four different applications by training several machine learning models using the silver standard dataset with several sample sizes and class imbalances and tested the performance using a gold standard dataset. Our evaluations on the four applications indicate the success of silver standard datasets in identifying a gold standard dataset. We conclude the study with evidence that noisy social media data can be utilized for weak supervision

DOI

https://doi.org/10.57709/30434727

File Upload Confirmation

1

Download

Share

COinS